How many bets before a record means anything
A 3-point edge takes roughly a thousand bets at prices around 2.50 to show up clearly. Why small samples lie, with the arithmetic and our own record.
On this page
Far more than feels reasonable. If you back selections at prices around 2.50 with a genuine 3-point edge, you need roughly a thousand bets before that edge stands clearly apart from luck. Under a hundred, your record is mostly noise, good or bad. That is not pessimism; it is what the arithmetic of a 40% chance does over a short run.
The arithmetic
A selection at 2.50 implies a 40% chance. Back enough of them and your win rate settles near the true chance, but the wobble around it shrinks only with the square root of the number of bets. One standard error on the win rate is the square root of 0.4 × 0.6 divided by the number of bets.
| Bets | One standard error on a 40% win rate | Two standard errors |
|---|---|---|
| 13 | 13.6 pts | 27.2 pts |
| 100 | 4.9 pts | 9.8 pts |
| 500 | 2.2 pts | 4.4 pts |
| 1,000 | 1.5 pts | 3.1 pts |
| 2,500 | 1.0 pts | 2.0 pts |
An edge of 3 points means your true chance is 43% where the price says 40%. Until two standard errors fall below 3 points, a run that looks like an edge could be a lucky 40% bettor, and a run that looks like no edge could be an unlucky 43% one. That line is crossed somewhere just past a thousand bets.
Our own record, read the same way
The BAI8 record has 79 settled calls. Thirteen of them carried a positive edge when published.
| Calls | Won | Expected wins (sum of our chances) | |
|---|---|---|---|
| All calls | 79 | 47 | 38.8 |
| Positive-edge calls | 13 | 4 | 5.8 |
Four wins against 5.8 expected is about one standard error short, which is ordinary. Forty-seven against 38.8 is ahead of expectation, which is also not proof of anything yet. We go through the thirteen edge calls one by one in our positive-edge calls: four from thirteen.
What to watch instead of the win count
- Wins against expected wins. Add up the stated chances. A record is only surprising relative to that sum.
- How the chances hold up in bands. Do the 40% calls win about 40% of the time over many calls? A score like the Brier score grades that; see Brier score explained for bettors.
- Your stake size. A sample this slow to settle is the reason to stake small and flat; the case for that is in staking without going broke.
What this does not tell you
It does not tell you whether you have an edge. It tells you how long you must wait before your results can say. The numbers above assume prices near 2.50 and independent bets; shorter prices settle faster, longer ones slower, and correlated bets (several picks on one evening's football) slower still.
18+. Bet only what you can afford to lose.
Common questions
How many bets do you need to know if you have an edge?
More than most people think. At prices around 2.50, a 3-point edge needs roughly a thousand bets before it stands two standard errors clear of zero.
Is a 13-bet record enough to judge a model?
No. With 13 bets at about 40% each, one standard error on the win rate is about 14 points, far wider than any realistic edge.
What should I look at instead of the win count?
Wins against expected wins, the gap to the closing price, and how well the stated chances match what happens over many calls.